multimodal LLM
Kimi K3
Moonshot AI's open-weight 2.8T MoE flagship; base for Cognition SWE-2 and available in GitHub Copilot. Video input on the card is inconsistent (register C4, not checked).
Levelmodel
Classmultimodal LLM
Parameters2.8T
Context1,048,576
Released16 Jul 2026
Opennessopen weights restricted
Runs onhyperscale cluster, multi-GPU node
Statusactive
Benchmark Results
vendor-reportedindependent
Arena Text leaderboardhuman preference arena
kimi-k3-max; top open modelindependent
1,488 ±5
1,2001,4001,600
Score in Elo; the axis starts at 1,200, as Elo has no zero. Higher is better.
rank 16, Arena Text overall
Artificial Analysis Intelligence Index v4.3.2composite index
Kimi K3 (Max)independent
44
03060
Score in index. Higher is better.
rank 3, Artificial Analysis open-weights view
maxindependent
44
03060
Score in index. Higher is better.
rank 31 of 205, AA leaderboard, 205 current models
BrowseCompagents and tool use
vendor-reported
91.2
050100
Score in %. Higher is better.
Kimi Code harness; 67.3 with mini-SWE-agentvendor-reported
67.5
050100
Score in %. Higher is better.
FrontierSWEcoding
Kimi Code harnessvendor-reported
81.2
04590
Score. Higher is better.
GPQA Diamondknowledge and reasoning
vendor-reported
93.5
050100
Score in %. Higher is better.
Humanity's Last Exam (HLE)knowledge and reasoning
with toolsvendor-reported
56
050100
Score in %. Higher is better.
without toolsvendor-reported
43.5
050100
Score in %. Higher is better.
OSWorld-Verifiedagents and tool use
vendor-reported
84.8
050100
Score in %. Higher is better.
ProgramBenchcoding
Kimi Code harnessvendor-reported
77.8
050100
Score in %. Higher is better.
SWE-Marathoncoding
Claude Code harnessvendor-reported
42
04590
Score. Higher is better.
Terminal-Bench 2.1agents and tool use
vendor-reported
88.3
050100
Score in %. Higher is better.
Kimi Code harnessvendor-reported
88.3
050100
Score in %. Higher is better.
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
Lineage
successorfinetune
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 6 models. Confidence and evidence are the notes' own.
R&D Topics
Identity
- Developer
- Moonshot AI
- Family
- Kimi
- Released
- 16 Jul 2026 sourcesinferred from page
- Weights published
- 27 Jul 2026
- Record checked
- 10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- MoE MoE, 896 experts, 16+2 active; KDA plus gated MLA hybrid attention
- Total parameters
- 2.8T sources
- Active parameters
- 104B
- Checkpoint parameters
- –
- Context length
- 1,048,576 tokens
- Input
- text, image
- Output
- text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights restricted
- Licence
- Kimi K3 License sources
- Access routes
- API, open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- hyperscale cluster, multi-GPU node
- Footprint
- 1.4 TB resident, 64+ accelerators (T2 multi-node cluster).
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Industry Use Cases
No industry use case links this model.
Links
DeveloperMoonshot AIstartup, China
Moonshot AIstartup, ChinaModelKimi K3Moonshot AI, multimodal LLM
Kimi K3Moonshot AI, multimodal LLMR&D TopicsEfficiency Architectures (MoE, Sparse Attention, Long Context)Open-Weight Frontier Releases as Research Substrate
Use CasesClinical LLM (Open Weights)HealthcareProcess Optimisation and Advanced Process ControlIndustrialTelecom Domain LLMs and BenchmarksTelecomUse cases of models in the same R&D topic; none is recorded for this model itself.
Who Runs It
PT Amman Mineral Nusa Tenggara (AMMAN)In production, 8 Jul 2026MIND IDResearch, 14 Oct 2025
Conch CementIn production, 23 Jun 2025
PT Kalbe Morinaga IndonesiaIn production, Oct 2024+2 moreDeployments of these use cases; none names this model.
PT Amman Mineral Nusa Tenggara (AMMAN)In production, 8 Jul 2026MIND IDResearch, 14 Oct 2025Conch CementIn production, 23 Jun 2025PT Kalbe Morinaga IndonesiaIn production, Oct 2024+2 moreDeployments of these use cases; none names this model.Models 3
Organisations 1
Sources
- Kimi K3 Tech Blog: Open Frontier Intelligence kimi.com https://www.kimi.com/blog/kimi-k3
- huggingface.co/api/models/moonshotai/Kimi-K3/commits/main https://huggingface.co/api/models/moonshotai/Kimi-K3/commits/main
- Moonshot Introduces 2.8T-Parameter Kimi K3, With Weights Due July 27 letsdatascience.com https://letsdatascience.com/news/kimi-k3-debuts-as-28-trillion-parameter-open-weight-model-cac7da05
- Kimi K3 (max) - Intelligence, Performance & Price Analysis | Artificial Analysis artificialanalysis.ai https://artificialanalysis.ai/models/kimi-k3
- Model Inference Pricing Explanation - Kimi API Platform platform.kimi.ai https://platform.kimi.ai/docs/pricing/chat
- Kimi K3 Open Weights Ship: 2.8T Parameters, 1.4 TB to Run rits.shanghai.nyu.edu https://rits.shanghai.nyu.edu/ai/kimi-k3-open-weights-ship-2-8t-parameters-1-4-tb-to-run/
- huggingface.co/moonshotai/Kimi-K3 https://huggingface.co/moonshotai/Kimi-K3
- Text Arena: LLM Rankings & AI Model Benchmarks arena.ai https://arena.ai/leaderboard/chat/text
- Supported AI models in GitHub Copilot - GitHub Docs docs.github.com https://docs.github.com/en/copilot/reference/ai-models/supported-models
- huggingface.co/moonshotai https://huggingface.co/moonshotai
- huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Kimi K3", https://aurion.semai.id/models/moonshot-ai-kimi-k3/, records of October 2026.